README.md

metadata

license: apache-2.0
task_categories:
  - image-classification
  - visual-question-answering
language:
  - en
size_categories:
  - 1K<n<10K

eval-hard-3500

A curated 3,500-sample hard evaluation benchmark for garment classification VLMs.

Format

JSONL with fields:

• image: path to garment image
• source: annotation source
• response: ground-truth JSON with 9 fields (type, color, pattern, neckline, sleeve_length, closure, brand, size, defect_type)

Usage

Used to evaluate multi-field structured JSON extraction from garment images. Models are scored on SBERT cosine similarity, NLI entailment, Levenshtein ratio, token F1, and weighted field scores.

Metrics

See eval_all_results.json for model comparison results on this benchmark.